Resurrecting this thread.
We're nearing the EOL for the non business version of Java SE 5.0
(business
edition will be available for quite a while - unless the new management
changes the plan) [1] .
When 5.0 goes out of service I'd propose upgrading OpenJPA to require
JDK
6.0 to compile. The compiled bytecode can be set to 1.5 if that's a
concern.
I'd prefer to have all the modules use jdk 6 to avoid some of the
headaches
we had in OpenJPA 1.0.x with supporting 1.4 but we can restrict it to
only
the ones that need it (persistence, persistence-jdbc) if that's more
amenable.
In addition we can set up a new integration module which runs a
subset of
tests with Java 5. It will be optional (since Java 5 won't be readily
available in 3 months), but at least we'd have some barometer for
whether
OpenJPA works in that environment. We'll have to do some classpath
swizzling
(like we did for 1.4 in the 1.0.x stream) but it *should* be possible.
Thoughts, objections, stuff I've missed?
[1] http://java.sun.com/products/archive/eol.policy.html
-mike
On Tue, Mar 31, 2009 at 2:29 PM, Michael Dick <michael.d.d...@gmail.com
wrote:
On Tue, Mar 31, 2009 at 2:07 PM, Pinaki Poddar <ppod...@apache.org>
wrote:
Hi Craig,
This also meets my needs for a stable platform to run a new
personality without the new Java 6 dependencies.
The current update in trunk runs a configuration that builds OpenJPA
libraries with JDK6 compiler. But other configuration compiles and
runs
our
test corpus with JDK5. I do not think we have a configuration that
compiles
OpenJPA with JDK6, compiles test cases with JDK5 and runs test cases
with
JDK5. May be we should create one. Such configuration will simulate
the
target JDK5 user environment with JDK6-compiled OpenJPA where the test
case
will play the equivalent role of user application.
(Mike/Jeremy, are you tuned to this channel?)
This is easier said than done. Depending on how strict one wants to be.
If
we rely on the compiler settings (source=1.5, target=1.5) when we
compile
the testcases then at worst we'd have to add a separate maven module
for
JDK5 testcases.
As we've seen in the past with JDK 1.4 this won't necessarily
suffice. We
may need to do some additional tweaking to put the 1.5 class
libraries on
the classpath, or (even more strict) we may need to rebuild with
maven's
JAVA_HOME set differently.
I'd be fine with the first approach as part of a normal build (provided
it
doesn't double execution time). Either of the later two would need
to be
optional (like we did with jdk 1.4).
mission statement for OpenJPA
"to the implementation of object persistence, including, but not
limited to, Java Persistence API, for distribution at no charge to
the
public;"
I fully agree and support this view. Compliance to a spec is a
necessary
but not sufficient condition for sustainable interest in a project of
OpenJPA's scope and breadth. Also one of the strongest feature of
OpenJPA is
its 'agnostic architecture' to promote the above charter.
As a group we will benefit if we keep the charter in mind and consider
possibilities to augment OpenJPA functionality that are beyond a
standard
specification.
I agree that the agnostic architecture is a strength of OpenJPA and one
that we can leverage to promote additional solutions in the ORM space.
That
said we are a JPA provider first and foremost and there are limits
to the
contortions that the "core" OpenJPA engine should make to support other
persistence frameworks. Especially those that have not been contributed
to
Apache.
To put it another way, our default behavior should be as JPA-like as
possible with the option for other frameworks to change the
configuration
to
suit their needs.
<snip>
3. If the above appears to be a worthwhile target scenario to
support, then the dynamic class construction approach perhaps can
prove useful than hand-coding JDBC 4 dependency.
4. We take a decision regarding these aspects by mid-April and
announce it to be effective from, say, mid-June. I am not keen on
exact duration of the prior notice but 2 months looked to be
reasonable.
Fair enough. My concern lies mainly with the dynamic class construction
and
the impact on performance. Introducing additional code path in order to
support a backleveled JDK seems wrong to me. Maybe I'm too anxious
to be
on
the bleeding edge.
-mike
<more message history snipped>